You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I'm not actually sure where this problem is happening, but it only happens on an EC2 box when targeting Sandy Bridge. If I recompile OpenBLAS to use Nehalem directly, I get rid of the warnings and the errors.
Why would a wrongly-configured BLAS make spawn.jl fail?
$ make testall
JULIA test/all
OpenBLAS : Your OS does not support AVX instructions. OpenBLAS is using Nehalem kernels as a fallback, which may give poorer performance.
OpenBLAS : Your OS does not support AVX instructions. OpenBLAS is using Nehalem kernels as a fallback, which may give poorer performance.
OpenBLAS : Your OS does not support AVX instructions. OpenBLAS is using Nehalem kernels as a fallback, which may give poorer performance.
OpenBLAS : Your OS does not support AVX instructions. OpenBLAS is using Nehalem kernels as a fallback, which may give poorer performance.
OpenBLAS : Your OS does not support AVX instructions. OpenBLAS is using Nehalem kernels as a fallback, which may give poorer performance.
OpenBLAS : Your OS does not support AVX instructions. OpenBLAS is using Nehalem kernels as a fallback, which may give poorer performance.
OpenBLAS : Your OS does not support AVX instructions. OpenBLAS is using Nehalem kernels as a fallback, which may give poorer performance.
OpenBLAS : Your OS does not support AVX instructions. OpenBLAS is using Nehalem kernels as a fallback, which may give poorer performance.
OpenBLAS : Your OS does not support AVX instructions. OpenBLAS is using Nehalem kernels as a fallback, which may give poorer performance.
From worker 3: * linalg2
From worker 5: * core
From worker 2: * linalg1
From worker 4: * linalg3
From worker 8: * strings
From worker 6: * keywordargs
From worker 7: * numbers
From worker 9: * collections
From worker 6: * hashing
From worker 6: * remote
From worker 6: * iobuffer
From worker 6: * arrayops
From worker 9: * simdloop
From worker 5: * blas
From worker 8: * fft
From worker 9: * dsp
From worker 8: * sparse
From worker 9: * bitarray
From worker 5: * random
From worker 5: * math
From worker 5: * functional
From worker 5: * bigint
From worker 7: * sorting
From worker 5: * statistics
From worker 8: * spawn
From worker 6: * backtrace
From worker 6: * priorityqueue
From worker 7: * arpack
From worker 6: * file
From worker 8: [stdio passthrough ok]
From worker 4: * suitesparse
From worker 5: * version
From worker 5: * resolve
From worker 4: * pollfd
exception on 8: ERROR: test failed: readall(@cmd "\$exename -f -e 'println(STDERR,\"Hello World\")'" .> @cmd "cat") == "Hello World\n"
The text was updated successfully, but these errors were encountered:
It's failing because that AVX warning is getting printed on every invocation of Julia, and the test in spawn.jl is checking that only "Hello World\n" gets printed (which is very important if you're redirecting the output of some big script to, e.g., a CSV file to be read later). I believe that the warning should be properly hidden by default in the new OpenBLAS release candidate (v0.2.9.rc2).
AFAIR, I also tested it with 0.2.9.rc2 (targeting Sandy Bridge, which is the default detected) and had the same issue, but let me retry.
Another solution would be to always default to Nehalem in paravirtualized EC2 boxes (since their Xen PV solution doesn't support AVX, while their HVM one does), but this is probably more relevant to OpenBLAS than Julia.
I'm not actually sure where this problem is happening, but it only happens on an EC2 box when targeting Sandy Bridge. If I recompile OpenBLAS to use Nehalem directly, I get rid of the warnings and the errors.
Why would a wrongly-configured BLAS make spawn.jl fail?
The text was updated successfully, but these errors were encountered: